Tactile Spatial Attention Enhances Gamma-Band Activity in Somatosensory Cortex and Reduces Low-Frequency Activity in Parieto-Occipital Areas

Introduction

Early studies investigating the mechanisms of spatial-selective attention in humans have used event-related potentials and have found increased sensory responses to attended compared with ignored stimuli (Groves and Eason, 1969; Hillyard et al., 1973). These effects have been corroborated in numerous experiments using both EEG/magnetoencephalography (MEG) and functional magnetic resonance imaging (MRI)/positron emission tomography (Brefczynski and DeYoe, 1999; Mangun et al., 2001) and have led to the conclusion that attention enhances the strength of the neuronal representations of attended stimuli, such that they have improved access to higher processing resources. Experiments using single-cell recordings in monkeys have revealed that multiple simultaneously presented stimuli compete for access to higher-level processing and that this competition is biased in favor of the attended stimulus. More specifically, Reynolds et al. (1999) presented two stimuli within the receptive field of a neuron in monkey area V4. When attention was directed to one of the stimuli, the neuron responded as if there were only the attended stimulus. What mechanisms increase the efficacy of an attended stimulus compared with an unattended stimulus?

It has been proposed recently that this modulation of input efficacy might be accomplished by a regulation of the synchronization among the involved neurons (Engel et al., 2001; Fries et al., 2001a,b; Salinas and Sejnowski, 2001). Synchronous input to a postsynaptic target neuron is known to have a greater impact than asynchronous input. A recent study in awake trained monkeys has indeed shown enhanced synchrony of stimulus-induced gamma-band oscillations in response to attended compared with unattended stimuli (Fries et al., 2001b). This is consistent with the finding of enhanced gamma-band activity for attended stimuli over parieto-occipital electrodes in the human EEG (Gruber et al., 1999). Furthermore, it has been shown that stimulus selection in binocular rivalry is associated with enhanced gamma-band synchrony in primary visual cortex (Fries et al., 1997, 2002).

Much less is known about how spatial-selective attention acts on individual neurons in the somatosensory system. One consistent finding from both animal (Hsiao et al., 1993; Burton et al., 1997; Burton and Sinclair, 2000a; Meftah et al., 2002) and human studies is that attentional effects are more robustly found in secondary somatosensory cortex (S2) and only to a much lesser degree in primary somatosensory cortex (S1) (Burton and Sinclair, 2000b; Hoechstetter et al., 2000). Furthermore, Steinmetz et al. (2000) recently found increased synchrony of neural responses in S2 when a monkey had to attend to tactile stimuli compared with when he had to attend to visual stimuli. However, it is not clear how results from this intermodal attention task relate to spatial-selective tactile attention. Also, this study did not analyze synchronization as a function of frequency, and therefore it is not clear how the enhanced synchronization described there relates to the enhanced gamma-band synchronization found with spatial-selective visual attention (Fries et al., 2001b).

Here, we investigated the effects of tactile attention on oscillatory neuronal synchrony and more specifically address the question whether spatial-selective tactile attention enhances gamma-band activity in the human somatosensory system. Therefore, we recorded MEG in healthy human subjects and used mechanical Braille stimuli in a tactile delayed-match-to-sample task.

Materials and Methods

Subjects. Subjects were nine adult volunteers (four males, five females; mean ± SD age, 25.1 ± 2.7 years) who were recruited primarily from the Radboud University Nijmegen. All subjects provided written consent according to institutional guidelines of the local ethics committee (Commissie Mensgebonden Onderzoek Region Arnhem-Nijmegen, The Netherlands), were reported to be free of mental and neurological illness, and were right-handed as assessed by the Edinburgh Handedness Inventory (Oldfield, 1971).

Stimuli. Tactile stimuli were presented to the left or right index fingers by means of two piezoelectrical Braille stimulators (Metec, Stuttgart, Germany) that were driven using custom-built electronic circuitry. Each Braille stimulator had a matrix of two columns by four rows of individually controllable pins, which could be raised and lowered with a 1-ms precision. The stimulus set consisted of five different dot patterns that were created with the constraint that, for each pattern, four pins were raised in clusters of at least two adjacent pins (in either horizontal or vertical direction; for details, see Fig. 1 A). Stimulation consisted of elevating the pin pattern by ∼2 mm, holding them elevated for 500 ms (sample stimulus) or 35 ms (test stimuli), and then lowering them again. While in the elevated position, the pins were stationary. The low currents necessary for driving piezoelectrical membranes created only small and short-lived artifacts. Those artifacts occurred at two times: (1) during the specification of the stimulus pattern in the electronic circuit driving the Braille cells, which happened 100 ms before actual stimulus delivery; and (2) during the actual delivery of the stimulus. These short (∼1 ms) artifacts can be seen in the time-domain averages in Figure 2. No artifacts were measured when the stimulus remained stationary (pins either elevated or lowered).

Task. The task was a spatial-selective tactile delayed-match-to-sample task that subjects performed while continuously fixating a central fixation cross. The temporal structure of the task is outlined in Figure 1 B. Each trial started with the presentation of the tactile sample stimulus (selected randomly from the stimulus set for each trial) to both fingers for 500 ms. Two hundred milliseconds after sample offset, a visual arrow was foveally presented for 500 ms and cued spatial-selective attention to either the right or left index finger. After the offset of the cue, there was no change in visual stimulation anymore (only stationary fixation cross). One second after the offset of the visual cue, the presentation of a series of up to eight test stimuli was started, and these stimuli were delivered unpredictably to either the right or the left index finger. The test stimuli were chosen randomly out of the set of five dot patterns (see above) and were presented each for 35 ms, and stimulus onsets were separated by 1000 ms. Short presentation time was chosen for several reasons: (1) to prevent subjects from making exploratory finger movements that might have confounded the attentional effects; (2) to avoid a separate off response during stimulus processing; and (3) to increase task difficulty and force subjects to really attend to the precued finger. The subjects' task was to press a button lying between both hands with both thumbs as soon as they recognized the occurrence of a target (i.e., the sample pattern) on the precued (attended) side. The button was constructed such that it only counted the response when pressed with both hands. No button should be pressed when a sample pattern was presented on the uncued (unattended) side. The trial stopped after presentation of a target, after the subject had pressed the button erroneously, or after eight stimuli had been presented. In ∼25% of the trials, no target was presented. Subjects were given feedback about the correctness of their behavior at the end of each trial. We restricted our analysis to correctly rejected nonsample patterns and therefore stimuli to which no motor response was given.

Procedure. After subjects were seated comfortably, electrodes and localization coils were attached (see below). Before the recording session, each pattern was presented 10 times in pseudorandom order to both the left and right index finger to familiarize the subject with the stimuli. A set of two experimental blocks with 16 trials each was given for training purposes. Throughout the recording session, auditory white noise was presented to the subjects through pneumatic earphones to mask the sound generated by the Braille cells.

The recording session consisted of 16 blocks of 16 trials each, resulting in ∼1 h of recording time. After the recording session, structural MRIs of each individual subject were made on a 1.5 T Siemens Sonata scanner (Siemens, Erlangen, Germany) using a standard T1-weighted sequence to enable reconstruction of each subjects head shape for the later described interpolation and source reconstruction procedures.

MEG recordings. Neuromagnetic activity was recorded using the whole-head 151 channel axial-gradiometer MEG system (Omega 2000; CTF Systems, Port Coquitlam, British Columbia, Canada) at the F. C. Donders Centre for Cognitive Neuroimaging. In addition, the electrocardiogram and vertical and horizontal electrooculograms (EOGs) were recorded through appropriate bipolar montages. The subject's head position relative to the MEG sensors was measured at the beginning and at the end of the session using three magnetic coils. One such coil was placed on the nasion. The other two coils were mounted onto the earpieces that held the tubes for pneumatic auditory stimulation such that the tubes ran through the middle of the magnetic coils. During the structural MRI scans, we used the same ear pieces but then with short tubes that contained a drop of vitamin E in place of the coils. This procedure allowed precise and reliable coregistration of the MEG and the MRI data. MEG data were low-pass filtered at 300 Hz and sampled continuously at a rate of 1200 Hz.

Data analysis. Data were analyzed using the Fieldtrip software package (http://www.ru.nl/fcdonders/fieldtrip/), a Matlab-based toolbox for the analysis of electrophysiological data that has been developed by our group. Data were checked for artifacts using a semiautomatic routine that helped detecting and rejecting eye blinks, muscle artifacts, and jumps in the MEG signal caused by the SQUID electronics. For each of these artifacts, an appropriate metric with specific sensitivity for the respective artifact was computed, and a rejection threshold was determined empirically. This was done separately for each subject, but subsequently, artifact rejection parameters for a given subject were fixed and applied automatically to its entire dataset. Adjustments to individual subjects' data are necessary because of differences in noise levels and in the signal-to-noise ratio of the EOG recordings.

The power-line artifact was removed using the following procedure. All signals had been recorded continuously for the entire duration of the recording session. For each time epoch of interest (and each recording channel), we first took a 10 s epoch out of the continuous signal with the epoch of interest in the middle. We then calculated the discrete Fourier transform (DFT) of the 10 s epoch at 50, 100, and 150 Hz without any tapering. Because the power-line artifact is of a perfectly constant frequency, the 10 s epoch contains integer cycles of the artifact frequencies, and all of the artifact energy is contained in those DFTs. We then constructed 50, 100, and 150 Hz sine waves with the amplitudes and phases as estimated by the respective DFTs and subtracted those sine waves from the 10 s epoch. The epoch of interest was then cut out of the cleaned 10 s epoch. Power spectra of the cleaned 10 s epochs demonstrated that all artifact energy was eliminated, leaving a notch of a bin width of 0.1 Hz (equivalent to s). The actual spectral analysis used the multi-taper method, with a spectral smoothing of ±2.5 Hz for the lower-frequency range and ±10 Hz for the higher-frequency range. Thus, the notch typically became invisible.

The artifact-free data were interpolated to a common sensor array template using a minimum-norm projection method (Knösche, 2002). Subsequently, planar gradients of the MEG field distribution were calculated using a nearest-neighbor method comparable with the method described by Bastiaansen and Knösche (2000).

Spectral analysis. We performed time-resolved spectral analyses using the multi-taper method, which offers optimal spectral concentration (similar to spectral smoothing) over the frequency range of interest, i.e., it allows to trade resolution in the frequency domain for reduced variance (Mitra and Pesaran, 1999). Two frequency ranges were analyzed separately with different window lengths and spectral concentrations. The lower-frequency band ranged from 5 to 40 Hz and was analyzed with a window length of 400 ms and a spectral concentration of ±2.5 Hz. The higher-frequency band ranged from 40 to 180 Hz and was analyzed with a window length of 200 ms and a spectral concentration of ±10 Hz.

Power spectra were separately computed for the horizontal and vertical planar gradients, and the resultant vector length of both was computed to obtain the power at that sensor location regardless of the orientation of the gradient. The variance of the power in each time–frequency bin was estimated using a jackknife procedure (Efron and Tibshirani, 1993). Subsequently, t statistics were calculated between all time–frequency bins and a common baseline (effective time windows, –400 to 0 ms for low frequencies and –250 to 0 ms for high frequencies). The resulting t values were transformed into z scores, averaged over local sensor groups, and pooled across subjects using the following formula: with z_i being the z score of the ith subject.

Those grand-average time–frequency z images showed clear spectral components and allowed the definition of time–frequency windows of interest. To test the statistical significance of the effects of attention on the power in the different time–frequency windows of interest, the Fourier transform was calculated using the multi-taper method with the time windows and spectral concentrations set to the respective time–frequency window of interest. For each subject, a z score was determined in the same manner as for the time–frequency analysis. To compare the two attention conditions, a paired t test was performed between the z scores of all subjects. The z scores are well normalized for intrasubject variance and therefore can be better compared across subjects than differences in absolute power.

To examine the phase relation of the measured oscillations to the onset of the stimulus, we repeated the same time–frequency analysis on the (non-bandpass filtered) evoked fields (see description below). Additionally, we calculated the time-resolved phase-locking factor by doing a time–frequency analysis with the same settings as described above, but followed by normalizing the complex Fourier spectra on each trial by their vector length and averaging those unit-length vectors across trials (Lachaux et al., 1999).

Source analysis. For the reconstruction of the neuronal sources of the spectral components that showed a statistically significant effect of stimulation or attention, we used an adaptive spatial filtering technique (Gross et al., 2001). Each subject's brain volume was divided into a regular 8 mm grid, and, for each grid location, a spatial filter was constructed. This filter has the property that it passes activity from the location of interest with unit gain but suppresses all other activity. The spatial filter w(r,f) is calculated according to the following formula: where L(r) is the forward model (lead-field matrix) at the location r of interest, C(f) is the cross-spectral density (CSD) matrix between all MEG signals at the frequency f, and λ is the regularization parameter. To compute the lead-field matrix, we used a multisphere model in which, for each sensor, a sphere was fitted to the head surface underneath that sensor. The head shape was derived from each individual subject's structural MRI and aligned to the MEG data.

To optimally capture the effect of interest, a time–frequency window was specified individually for each subject based on the stimulus-induced power changes. A window of the same extent in time and frequency was also placed into the baseline period. For some of the source analyses, individual subjects were excluded because it was either impossible to find a proper baseline or because the signal-to-noise ratio was not sufficient. This will be described in detail in Results. The CSD matrix was calculated directly for the entire time window of interest using multi-tapers. Noise in the CSD estimate was minimized by choosing one frequency bin and adjusting the number of tapers (and thereby integration in the frequency domain) such that the entire frequency range of interest was captured. The time windows were chosen to represent an integer number of cycles of an oscillation of the center frequency, and the spectral concentration was chosen to be an integer multiple of the Rayleigh frequency of the selected window length.

Consecutively, the power p was estimated for each grid location r and separately for the two conditions that were compared (stimulus effect: baseline and stimulation; attention effect: attended and nonattended) according to the following formula: To quantify the effects of tactile stimulation and tactile attention on oscillatory activity, we computed a voxel-wise t statistic. To this end, the cross-spectral density matrices were calculated for each single trial and then jackknife averaged. The spatial filter was constructed for each individual jackknife subsample, resulting in an estimate of the mean and the variance of power at each grid location and for each condition. A t statistic was then determined for the difference in power between the respective conditions. The t values were subsequently z transformed. To analyze the effect of stimulation, we chose, for each spectral component separately, the attention condition that provided the highest signal-to-noise ratio. Because for one comparison (effect of attention on alpha activity), the baseline levels were systematically different for the attended compared with the unattended condition, we used for this comparison not a simple t test but the following contrast: (AttendedPostStim – AttendedPreStim) – (UnattendedPostStim – UnattendedPreStim).

Using SPM2 (http://www.fil.ion.ucl.ac.uk/spm), the individual anatomical MRIs and the corresponding statistical maps were spatially normalized toward the International Consortium for Brain Mapping template (Montreal Neurological Institute, Montreal, Quebec, Canada). The individual spatially normalized statistical maps were subsequently pooled to obtain a fixed-effect statistic using the same formula as for the sensor data.

Additionally, we performed a source reconstruction of the first two peaks of the somatosensory-evoked field to investigate the overlap of the frequency domain results with those of the well described generators of early evoked activity. To this end, we also used an adaptive spatial-filtering approach, similar to the one described above, but with the cross-spectral density matrix replaced by its time-domain equivalent, the covariance matrix between all MEG signals (Van Veen et al., 1997). This was done separately for a prestimulus and a poststimulus time window, and subsequently, the same statistical and cross-subject pooling procedures were applied as for the frequency domain data.

Time-domain analysis. The same artifact-corrected raw data as used for the spectral analysis were low-pass filtered (cutoff frequency set to 40 Hz), jackknife averaged, and baseline corrected (baseline from –100 ms to stimulus onset), interpolated to the common sensor array and planar gradiometer transformed. The obtained realigned planar gradients were then grand averaged to obtain the average topographies. The subsequent analysis proceeded similarly to that of the spectral analysis: mean and variance were estimated based on the jackknife repetitions, and subsequently, a t statistic was calculated for the difference between each sample measured after stimulus onset and the mean amplitude during the baseline period. The t values were then z transformed and pooled across subjects. Finally, the time courses for the attended and unattended conditions were subjected to a paired t test to test for statistically significant differences between the conditions. However, because an analysis of time-locked effects was not the focus of the present work, we only tested for the presence of such effects.

Results

Discussion

We investigated the effects of tactile stimulation and spatial-selective tactile attention on oscillatory neuronal activity in humans. We found an early stimulus-induced, but not stimulus-locked, gamma-band response (60–95 Hz). The onset latency, the strong contralateral dominance, and the source reconstruction strongly suggest an origin in primary somatosensory cortex. This is consistent with other recent studies that have reported gamma-band activity in the human primary somatosensory cortex recorded with MEG (Hirata et al., 2002; Gaetz and Cheyne, 2003; Ihara et al., 2003). Those previous studies, however, used electrical median nerve stimulation in contrast to mechanical stimulation as used here. MEG and EEG activity to electrical stimuli have different characteristics, such as the presence of very short latency responses in early somatosensory areas (e.g., N20m–P35) (Simôes et al., 2003) that are absent in our data and those of others using mechanical stimulation of the skin (Simôes et al., 2001; Braun et al., 2002).

We found that stimulus-induced gamma-band activity in somatosensory cortex is strongly enhanced and prolonged by spatial tactile attention. Source reconstruction revealed colocalization of this attentional enhancement with the effect of stimulation in contralateral S1. This result is in contrast to findings from other electrophysiological studies assessing spike rates or event-related potentials/fields that have generally found attentional effects to be stronger in S2 and rather weak or absent in S1 (Burton and Sinclair, 2000a; Hoechstetter et al., 2000) (but see Braun et al., 2002). However, functional imaging studies, relying on hemodynamic signals, did reveal attention effects in S1 (for review, see Burton and Sinclair, 2000b). It is by now well established that the hemodynamic response is more tightly correlated with synchronized gamma-band activity than with neuronal firing rates (Logothetis et al., 2001; Niessing et al., 2005). Therefore, one possible explanation for the apparently conflicting results from the different approaches is that attention synchronizes gamma-band oscillations in primary somatosensory cortex, which translates into enhanced spiking activity at later processing stages. We will further discuss this point at the end of Discussion.

The attentional enhancement in this study commenced ∼100 ms and lasted until ∼500 ms after stimulus onset. The effect was found in the data of eight of nine subjects, and the time courses of gamma-band activity were highly overlapping for stimulation of either hand. Among the early responses in the somatosensory system, the attentional effect on the induced gamma-band activity was stronger and more consistent than attentional effects on oscillatory activity in other frequency bands or on stimulus-locked components. This suggests that gamma-band oscillations play an important functional role in the early stages of the somatosensory system for processing behaviorally relevant stimuli. We hypothesize that gamma-band synchronization in primary somatosensory cortex is instrumental in communicating with higher somatosensory areas.

Although tactile stimulation induced gamma-band activity, it also led to the well known suppression of the somatosensory mu (≈8–15 Hz) and beta (≈15–25 Hz) activity (Crone et al., 1998; Pfurtscheller et al., 2002) and the subsequent beta rebound (≈15–25 Hz) (van Burik and Pfurtscheller, 1999). Consistent with some of the previous studies, mu and beta suppression are widespread in both their spectral distribution (ranging from 7 to 40 Hz) and spatial distribution, peaking over somatosensory/sensorimotor cortex contralateral to the stimulated hand but also reaching to ipsilateral sensorimotor cortex and bilateral parietal and occipital regions. Stimulus-induced modulations of mu- and beta-band activity were of very high signal-to-noise ratio. Nevertheless, there was no significant modulation of the somatosensory mu and beta suppression with spatial-selective attention. This suggests that the respective desynchronization phenomena are not directly involved in the mechanisms underlying the attentional selection of behaviorally relevant stimulus information. Rather, they appear to reflect an automatic response of the sensorimotor circuit to afferent stimulation that occurs independently of processing demands. In contrast, there was a significant reduction of the beta rebound with spatial-selective attention. This attentional effect started ∼500 ms after onset of the nontarget stimuli, whereas the average reaction time to the target stimuli was in the order of 510 ms. Therefore, a direct functional role of the attentional modulation of the beta rebound for sensory processing or decision making in the somatosensory system has to be considered unlikely.

An interesting attentional effect was found for the occipital alpha-band (7.5–12.5 Hz) and beta-band (13.5–26.5 Hz) activity. Attended tactile stimuli resulted in a deeper and longer-lasting suppression of occipital alpha and beta-band activity than unattended tactile stimuli (Eimer, 2000; Worden et al., 2001; Kennett et al., 2003). Furthermore, tactile stimulation did not only alter ongoing oscillatory activity in occipital cortex but also resulted in stimulus-locked activation over occipital cortex. Evoked fields showed a clear spatial peak over medial occipital cortex, with an onset latency of ∼350 ms. There was no consistent attentional modulation of this component, in contrast to the strong attentional modulation of the stimulus-induced occipital alpha and beta suppression. This suggests different functional mechanisms behind those two processes.

Previous hemodynamic studies have shown that tactile stimulation can activate various parts of visual cortex in blind (Sadato et al., 1996) and normally sighted (Zangaladze et al., 1999; Amedi et al., 2002) human subjects. In principle, this might be related to direct input from heteromodal areas (Falchier et al., 2002; Schroeder and Foxe, 2002) or to mechanisms underlying visual imagery that are not yet fully understood (Zhang et al., 2004). The tactile evoked fields that we found over occipital areas showed a late onset. We therefore consider it unlikely that they were attributable to direct input from early somatosensory areas.

One potential concern is that physical stimulation might not have been identical for the two attention conditions because attention might have invoked motor programs to optimize haptic sensing. For example, subjects might have made exploratory movements or pushed the attended finger slightly stronger onto the Braille cells than the unattended finger. For several reasons, we are convinced that this is not the case. First, stimulation was very short, lasting only 35 ms and thereby excluded the possibility that subjects engaged in active exploration of the stimuli. Second, the time courses of gamma-band activity and evoked fields are identical in the two attention conditions until ∼100 ms after stimulus. Third, pushing stronger on the attended side does not seem to be a good strategy because the Braille stimulators use only weak forces to drive the pins and are designed to give optimal sensation when the finger is resting relaxed on them. The pins give a diminished tactile sensation when pushed on or when the finger is lifted off from them. Thus, we can almost certainly rule out that differences in physical stimulation account for the attentional effects reported here.

The mechanisms underlying the observed attention effects are still unknown, and we can only speculate at this point. Neuronal networks have typically been found to show enhanced gamma-band activity during activation (Tiesinga et al., 2001; Cunningham et al., 2003). Thus, the attentional enhancement of gamma-band activity observed here might be attributable to excitatory input from, for example, a frontoparietal network responsible for the allocation of spatial attention (Corbetta and Shulman, 1998). Alternatively or in addition, top-down input might act through neuromodulators. Acetylcholine seems to support the generation of local gamma-frequency synchronization (Munk et al., 1996; Fellous et al., 2001; Rodriguez et al., 2004) and also the propensity of neurons to be entrained by rhythmic input (Schreiber et al., 2004). Such modulatory input might be of corticocortical or subcortico-cortical nature, but it would have to possess at least a crude spatial selectivity to explain the present results. There might also be top-down input to the neurons processing the unattended stimulus. This input might activate local inhibitory networks to decrease the activation of those neurons. Increased activation of inhibitory networks leads to enhanced beta-band synchronization (Jensen et al., 2005).

The effect of selective attention on stimulus-related gamma-band activity was mainly restricted to somatosensory cortex. Many previous studies have reported the activation of other areas during the performance of attention tasks, in particular areas of the frontoparietal attention network (Corbetta and Shulman, 1998). Those different findings are fully compatible with each other. The attention network is typically activated during cue processing and during the allocation of attention (Giesbrecht et al., 2003). In our paradigm, the investigation of this network would have required the comparison between task-related activity and a precue baseline. Rather, we focused here on the effects of selective attention on stimulus-related activity. To this end, we performed direct comparisons between the two attention conditions for time–frequency windows with the strongest stimulus-related activations. The analysis of oscillatory synchronization in the attention network and its interaction with oscillatory synchronization in sensory areas is of great interest but goes beyond the scope of this study.

To summarize, spatial-selective tactile attention enhances and prolongs gamma-band activity in primary somatosensory cortex. This will likely enhance the impact of the activated neurons on their postsynaptic targets, e.g., in secondary somatosensory cortex (Salinas and Sejnowski, 2001; Fries, 2005). Intracellular recordings in visual cortex in vivo have revealed recently that postsynaptic target neurons primarily respond to rapid fluctuations in their synaptic input (Azouz and Gray, 2003). Thus, a neuronal group providing input may enhance its impact through either synchronizing spikes in the gamma band or producing more spikes in the gamma band. In both cases, the MEG power in the gamma band would be enhanced. Investigations using microelectrode recordings of spikes and local field potentials showed that local field potential power can be enhanced in the absence of changes in firing rates but the presence of changes in local synchronization (Fries et al., 2001b). Increases in gamma-band activity in S1 will likely lead to generally increased activation of S2 and thereby might explain some of the attentional effects found in S2. Furthermore, it is conceivable that attention generally renders the reciprocal information exchange between different parts of the neural network, such as S1 and S2, more effective through a stronger oscillatory coupling (Fries, 2005; Schoffelen et al., 2005). These questions will be explicitly addressed in future experiments.